assessment of the impact of a personalised nutrition ... · reml restricted maximum likelihood snp...
TRANSCRIPT
ASPIRE – SAP – version 3 of October 19th, 2018 1
ASPIRE-DNA
Assessment of the Impact of a Personalised Nutrition Intervention in Impaired
Glucose Regulation
Statistical analysis
Author: Pierre Debeaudrap
Version 3 of October 19th, 2018
Version Date Summary of changes
1 July 6th, 2018 Reviewed by CG
2 September 13th, 2018 Additional information on the variables
measurement
HbA1c included as endpoint (following
change in the protocol)
Section exploratory analysis (leptin and
comparison of the exploratory arm)
completed
Various minor correction including
editing
3 October 19th, 2018 Final version
ASPIRE – SAP – version 3 of October 19th, 2018 2
Table of contents Investigating team --------------------------------------------------------------------------------------- 5
Signature for approval --------------------------------------------------------------------------------- 5
Background ----------------------------------------------------------------------------------------------- 6
Study objectives ------------------------------------------------------------------------------------------ 7
1.1 Primary objectives --------------------------------------------------------------------------------------- 71.2 Secondary objectives ------------------------------------------------------------------------------------ 71.3 Exploratory objectives ---------------------------------------------------------------------------------- 7
Study design ---------------------------------------------------------------------------------------------- 8
Definitions ------------------------------------------------------------------------------------------------- 8
Endpoints -------------------------------------------------------------------------------------------------- 8
1.4 Primary endpoint ---------------------------------------------------------------------------------------- 81.5 Secondary endpoints ------------------------------------------------------------------------------------ 9
Study population ---------------------------------------------------------------------------------------- 10
Statistical considerations ------------------------------------------------------------------------------ 11
1.6 Analysis population ------------------------------------------------------------------------------------ 111.7 Data description ---------------------------------------------------------------------------------------- 11
1.7.1 Recruitment and follow-up ----------------------------------------------------------------------- 111.7.2 Data summary -------------------------------------------------------------------------------------- 111.7.3 Data review and exploration of outliers and problematic distribution ---------------------- 13
1.8 Baseline characteristics and description of the study population ---------------------------- 131.9 Analyses of quantitative data ------------------------------------------------------------------------ 161.10 Repeated measures analyses ----------------------------------------------------------------------- 161.11 Non-parametric model ------------------------------------------------------------------------------ 171.12 Missing data ------------------------------------------------------------------------------------------- 17
Analysis of the primary endpoint ------------------------------------------------------------------- 17
1.13 Derivation ---------------------------------------------------------------------------------------------- 171.14 Planned analysis -------------------------------------------------------------------------------------- 18
Secondary endpoints ----------------------------------------------------------------------------------- 19
1.15 Variables related to glucose regulation: --------------------------------------------------------- 191.16 Variables related to anthropometric measurements ------------------------------------------ 201.17 Variables related to lipid profile ------------------------------------------------------------------ 21
ASPIRE – SAP – version 3 of October 19th, 2018 3
1.18 Variables related to clinical markers ------------------------------------------------------------- 211.19 Variables related to nutrition ---------------------------------------------------------------------- 21
Exploratory analyses ----------------------------------------------------------------------------------- 22
Additional information -------------------------------------------------------------------------------- 22
1.20 Significance level -------------------------------------------------------------------------------------- 221.21 Statistical software ----------------------------------------------------------------------------------- 22
References ------------------------------------------------------------------------------------------------ 24
ASPIRE – SAP – version 3 of October 19th, 2018 4
Abbreviations
BMI Body Mass Index
CRF Case report form
DBP Diastolic Blood pressure
DNA Deoxyribonucleic acid
FFQ Food Frequency Questionnaire
FPG Fasting Plasma Glucose
HDL High-density lipoprotein
HOMA Homeostasis model assessment of beta cell function
IFT Impaired fasting glucose
IGR Impaired Glucose regulation
IGT Impaired glucose tolerance
IQR Inter quartile range
IRI Insulin resistance index
ITT Intention to treat
LDL Low-density lipoprotein
LMM Linear mixed model
LTFU Loss to follow up
OGTT Oral glucose tolerance test
PG120 Plasma glucose at 12 minutes
QQ plot Quantile – quantile plot
REML Restricted Maximum Likelihood
SNP single nucleotide polymorphisms
SBP Systolic blood pressure
ASPIRE – SAP – version 3 of October 19th, 2018 5
Investigating team
Chief Investigator Prof. Nick Oliver
Co-Investigator Prof. Chris Toumazou
Co-Investigator Dr. Maria Karvela
Trial Manager Dr. Caroline Golden
Statistician Dr. Pierre de Beaudrap
Dietitian Ms. Natalie Bosnic
Project Officer Mrs. Maria Eze
Research Nurse Ms. Judith Bedzo-Nutakor
Laboratory Scientist Dr. Sara de Mateo Lopez
Laboratory Scientist Mr. Khalid Mirza
Laboratory Scientist Ms. Francesca Cavallo
Operations Director, DnaNudge Dr. Tsz Kin Hon
Signature for approval
Name Role Signature Date
Author
Dr Pierre Debeaudrap Statistician 19/10/2018
Approved by
Prof. Nick Oliver Chief investigator
22/10/2018
Dr Caroline Golden Trial manager
22/10/2018
ASPIRE – SAP – version 3 of October 19th, 2018 6
Background
Diabetes is amongst the most common long term conditions, with the number of people
affected worldwide quadrupling from 108 million in 1980 to 422 million in 20141. Its
prevalence in people over 18 years of age has risen from 4.7% in 1980 to a staggering 8.5% in
2014. In 2012, there were 1.5 million deaths as a direct result of diabetes, making it the 8th
leading cause of death amongst both sexes, and the 5th leading cause of death amongst
women. There were a further 2.2 million deaths as a result of complications due to higher-
than-optimal glucose levels2. In 2013, 6% of the UK adult population (2.7 million people)
were diabetic3, 90% of whom had type 2 diabetes4. A further 5 million people were estimated
to be at high risk of developing type 2 diabetes (NHS n.d.). This has led to a cost of £8 billion
per year to the NHS, 80% of which is due to diabetes-related complications such as
cardiovascular disease, amputations, renal failure and sight loss5.
The potential for lifestyle interventions to prevent type 2 diabetes in high-risk patients has
been well demonstrated through various clinical trials6,7. Yet there is important variation in
the response between individuals who receive lifestyle interventions because of behavioural
factors such as personal motivation and adherence but also because of genetic factors8. When
personalised, the lifestyle interventions could achieve greater effectiveness by optimising and
adapting the intervention components to individuals’ goals and characteristics as well as
enhancing motivation to adhere to dietary advice. There is a need to determine the most-cost
effective personalised interventions and how they could be applied.
The ASPIRE study aims to assess the use of genetic data to guide personalised nutrition in
people with impaired glucose tolerance and to demonstrate the impact of personalised
nutrition on glucose metabolism, anthropometry, non-glucose metabolism and lifestyle.
In this trial, it is assumed that DNA-based dietary advice is more effective in improving
glucose metabolism in pre-diabetic participants compared to the advice received as part of
standard care for pre-diabetic individuals. A DNA-based diet is also assumed to improve
anthropometric measurements, lipid profile and blood pressure in pre-diabetic participants. If
participants from the intervention arm are found to have lower fasting plasma glucose (FPG)
compared to patients from the control arm, this will provide evidence that DNA-based dietary
advice is an effective way to prevent diabetes.
ASPIRE – SAP – version 3 of October 19th, 2018 7
This document details the planned analyses and endpoints derivation for the ASPIRE-DNA
study.
Study objectives
1.1 Primary objectives
§ To compare differences in the impact of a DNA-based diet and standard care
in improving glucose regulation in pre-diabetic individuals
1.2 Secondary objectives
§ To assess the ability of a DNA-based diet to improve the macro- and micro-
nutrient profile of pre-diabetic participants.
§ To assess the effect of a DNA-based diet on clinical markers including
cholesterol, body composition and blood pressure, as they relate to diabetes
risk.
§ Where applicable, to assess changes in anthropometric measurements as a
result of following a DNA-based diet, where they relate to diabetic risk
reduction.
1.3 Exploratory objectives
§ To assess the impact of providing DNA-based dietary guidelines via the
DnaNudge App, on improving glucose regulation in pre-diabetic individuals
§ To assess the impact of providing DNA-based dietary guidelines via the
DnaNudge App, on the secondary outcomes outlined above.
§ To explore the utility of an app as a delivery mechanism for DNA-based
dietary advice.
§ To examine the changes in leptin measurements via saliva samples during the
study.
ASPIRE – SAP – version 3 of October 19th, 2018 8
Study design
Open label randomised control trial. Participants will be randomised 1:1:1 to a) the control
arm (n = 60) in which they will receive general health guidelines according to the NICE
guidelines, b) the intervention arm (n = 60) in which they will receive DNA-based health
guidelines via a genetic report, or to c) the exploratory arm (n = 60) in which they will receive
DNA-based guidelines via the DnaNudge App.
The primary analysis will focus on the comparison between intervention arm and control arm.
Definitions
Impaired glucose regulation: impaired fasting glucose (IFG) or impaired glucose tolerance
(IGT).
Impaired fasting glucose:
6.1mmol/L ≤ Fasting plasma glucose (FPG) ≤ 6.9mmol/L AND 2 hr plasma glucose: < 7.8
mmol/L
Impaired glucose tolerance:
- FPG ≥7mmol/L AND 7.8mmol/L ≤ 2-hour plasma glucose < 11mmol/L after glucose
load
- HbA1c between 42 and 47 mmol/mol (6.0–6.4%)
Endpoints
1.4 Primary endpoint
The primary endpoint for the ASPIRE trial is the difference in FPG at week 6 between the
intervention arm that receives DNA-based diet advice and the control arm. This outcome has
been chosen because it is the most sensitive indicator of impaired glucose regulation.
Endpoints assessment will not be blind to the study intervention arm allocated.
ASPIRE – SAP – version 3 of October 19th, 2018 9
1.5 Secondary endpoints
Secondary outcomes include differences at weeks 6, 12 and 26 between arms in the cross
sectional values and changes from baseline of the variables listed below.
• Variables related to glucose regulation:
- FPG at other time points
- Plasma glucose at 120 minutes (P120) following 75g OGTT
- Insulin resistance index (IRI) computed using the Homeostasis model assessment
(HOMA) of beta cell function
- c-peptide at 120 minutes following 75g OGTT
- HbA1c
• Variables related to anthropometric measurements
- Body weight
- Body Mass Index (BMI)
- Lean mass
- Fat mass
- Waist circumference
• Variables related to lipid profile
- Total cholesterol
- Fasting triglycerides
- Low-density lipoprotein (LDL) cholesterol
- High-density lipoprotein (HDL) cholesterol
• Variables related to clinical markers
- Systolic Blood pressure (SBP)
- Diastolic blood pressure (DBP)
• Dietary intake
- Total energy intake
- Carbohydrate intake
- Fat intake
- Saturated fat intake
- Salt intake
ASPIRE – SAP – version 3 of October 19th, 2018 10
- Vitamin D
- Vitamin B6
- Vitamin B12
Study population
The study participants will be adults with impaired glucose regulation and fulfilling the
following inclusion / exclusion criteria:
• Inclusion criteria
- Provision of signed and dated informed consent form
- Adults over 18 years of age
- Impaired glucose regulation including IFG and IGT by FPG, OGTT or HbA1c
criteria
- Access to smartphone with an operating system of iOS 8.0 or above, or Android
4.0 or above.
• Exclusion criteria
- Diabetic
- Pregnant or planning pregnancy
- Breastfeeding
- Enrolled in other clinical trials
- Have active malignancy or under investigation for malignancy
- Severe visual impairment
- Reduced manual dexterity
- Use of psychiatric, anti-diabetic, and/or weight loss medication, and/or oral
steroids
- Bariatric surgery
- History of illnesses that could interfere with the interpretation of the study results
(e.g. HIV, Cushing syndrome, chronic kidney disease, chronic liver disease,
hyperthyroidism, hereditary fructose intolerance, alcohol or substance abuse)
- Unable to participate due to other factors, as assessed by the Chief Investigator
ASPIRE – SAP – version 3 of October 19th, 2018 11
Statistical considerations
1.6 Analysis population
- Intention to treat population (ITT)
Primary analyses will be conducted following the intention to treat (ITT) principle. All
participants who are randomised into the study will be included in the ITT analysis and the
analysis is conducted according to the randomised treatment arm.
- Per protocol population
In addition, the primary endpoint will be analysed with the Per Protocol population (PPP),
which consists of those subjects in the ITT population who complete the study with no
significant deviations from the planned protocol procedures. The exclusion will be as follows:
- Participants without impaired glucose regulation before inclusion [see definition
page 6]
- Significant protocol deviation such as non respect of inclusion / exclusion criteria
- Pregnancy during follow-up
- Participants loss to follow-up or who withdrew consent
1.7 Data description
Data will be collected as described in protocol version 0.18 of 16/07/2018 and managed as
described in the data management system version 0.5 of 01/10/2018. Data will be collected in
electronic case report form (eCRF) completed by investigators. All data will be on the virtual
machine for the medical statistician to work with, the data will be in the form of data
extractions (i.e. not the CRFs). If the original CRFs are needed, they will be provided in a
read only format.
1.7.1 Recruitment and follow-up
Date of first and last study inclusion and of the last visit will be reported. A flow chart will be
used to describe participants’ flow. It will include the following information: number of
persons screened, eligible, included, assessed at week 6 visit (see Figure 1). In addition, the
number of participants screened, recruited, followed until study completion, lost to follow-up
and who were withdrawn will be presented by month and intervention group in a table.
1.7.2 Data summary
Collected data will be summarised without consideration for treatment group as follows:
ASPIRE – SAP – version 3 of October 19th, 2018 12
- Quantitative data: maximum and minimum values, median and interquartile range
(IQR);
- Qualitative data: count (n) and proportion (%);
In addition, for each variable, the proportion of missing data will be reported. This
information will be reported using the following output table (Table 1 and 2).
Figure 1. Example of flowchart
Table 1. Example of output table for main continuous variables
Variable Label N Missing Min Max P25 P75 median
Table 2. Example of output table for main categorical variables
Variable Label N Missing n % Cumulative
n %
ASPIRE – SAP – version 3 of October 19th, 2018 13
1.7.3 Data review and exploration of outliers and problematic distribution
The distribution of the main variables (e.g., FPG, triglyceride, cholesterol) will be assessed
using published reference data. These distributions will be compared against a normal
distribution using a quantile – quantile (QQ) plot. Skewness and kurtosis will be computed
and additional assessment of normality may be conducted using the Kolmogorov-Smirnov
test. Boxplot will be used to detect outliers.
Variables with outliers, missingness and/or problematic distribution will be reviewed with the
some/all (where applicable) of the following members of the Trial Team:
• Ms. Natalie Bosnic (Dietitian)
• Ms. Judith Bedzo-Nutakor (Research Nurse)
• Mrs. Maria Eze (Project Officer)
• Dr. Caroline Golden (Trial Manager)
• Prof. Nick Oliver (Chief Investigator)
1.8 Baseline characteristics and description of the study population
Table 1 below lists the baseline variables of interest and presents the methods of derivation.
The baseline variables will be summarised for each treatment arm using the mean with
standard deviation (sd) for continuous data and count (n) with proportion (%) for categorical
data. Statistical testing of between group imbalance in the baseline covariates will not be
performed unless required for regulatory or publication reasons9. A tabular presentation will
be used to display the results.
ASPIR
E – SAP – version 3 of O
ctober 19th, 2018
14
Table 1. Baseline variables
Description
Definition
Type
Unit
Coding /derivation
CR
F
Age
N
um
Year
Num
ber of years from
date of birth (months :
year)
Anonym
ised
Random
isation log
Sex
N
(%)
Screening
enrolment
log
HbA
1c
Num
M
mol/m
ol
or %
V
isit2-CR
F
Random
isation group Intervention
group
assigned at randomization
A
nonymised
Random
isation log
Height
m
V
isit3-CR
F
Weight
Kg
V
isit3-CR
F
Body m
ass index (BM
I)
K
g/m2
V
isit3-CR
F
BM
I>30
N
(%)
V
isit3-CR
F
Fat body mass
Kg
V
isit3-CR
F
Lean body mass
Kg
V
isit3-CR
F
Waist circum
ference
cm
Visit3-C
RF
Systolic BP
mm
Hg
V
isit3-CR
F
Disatolic B
P
m
mH
g
Visit3-C
RF
ASPIR
E – SAP – version 3 of O
ctober 19th, 2018
15
Pulse
bpm
Visit3-C
RF
Glucose t0
Plasma glucose at tim
e 0
after 75g glucose load
Num
m
M/L
V
isit3-CR
F
Glucose t120
Plasma
glucose at
120’
after 75g glucose load
Num
m
M/L
V
isit3-CR
F
Insulin t120 Plasm
a Insulin
at 120’
after 75g glucose load
IU
/L
Visit3-C
RF
C-peptide t0
C-peptide at t0 after 75g
glucose load
m
M/L
V
isit3-CR
F
Total cholesterol
m
M/L
V
isit3-CR
F
HD
L-cholesterol
m
M/L
V
isit3-CR
F
Non-H
DL cholesterol
mM
/L
Visit3-C
RF
LDL cholesterol
mM
/L
Visit3-C
RF
Fasting TG
mM
/L
Visit3-C
RF
ASPIRE – SAP – version 3 of October 19th, 2018 16
1.9 Analyses of quantitative data
Continuous outcomes will be compared between groups using the modified analysis of
covariance (ANCOVA)10. This approach has been chosen because it is more powerful than
other alternatives such as t-test11.
A linear regression model will be used to analyse the association between post-intervention
values of the outcomes with intervention arm adjusting for their baseline value, participant
age and sex.
Let yit denote the outcome value for individual i at time t, xi her covariates (e.g. sex, age) and
di her intervention arm. The ANCOVA model is given by:
An extension of the basic ANCOVA model will consider an interaction between the
intervention indicator and the baseline values to test the assumption that the treatment effect
varies with baseline status.
In addition, changes in continuous outcomes from baseline will be analysed using a linear
regression model of the difference between the outcome value at visit and its baseline value.
Using the same notation, the change model is
Covariates will be centered in the analyses. Robust linear regression lmrob from the R
package robustbase will be used for the ANCOVA analysis12 unless specified otherwise. It
uses the MM-estimator to estimate the regression coefficient accounting for departure from
normality of the outcomes.
Assessment of models’ goodness of fit will include careful examination of residuals using
graphical methods (scatter plot, boxplot and QQ plot) and detection of outliers. In addition,
equality of the post-treatment variances will be tested using Levene’s test.
1.10 Repeated measures analyses
Linear mixed model (LMM) will be used to analyse changes over time of continuous
outcomes. Using the same notation (yit for the outcome value for individual i at time t, xi the
covariates, and di for the intervention arm), the following model will be used to model the
change in yit
ASPIRE – SAP – version 3 of October 19th, 2018 17
LMM will be estimated using the REML method13.
This model can be extended to account for non linearity in the time trend and for an
interaction between time trend and intervention arm as follows:
The assessment of the model’s goodness of fit will be performed through graphical
examination of the within-group residuals and random effects. It will include scatter plots of
standardised residuals against fitted values as well as QQ plot and boxplot of standardised
residuals and of random effects14. The homoscedastic normal errors assumptions will be
tested using the Shapiro-Wilks test and White’s test of homoscedasticity of error variances.
1.11 Non-parametric model
For variables showing strong departure from normality, the non-parametric Mann Whitney
Wilcoxon rank sum test (two groups) or the Kruskal and Wallis (to compare the three arms)
will be used.
1.12 Missing data
Missing data will be imputed using the multiple imputation with chained equation (MICE)
method using the other variables including outcomes variables if there is no evidence against
the assumption of data missing at random1517. Missing at random refers to the fact that all
factors related to the probability of a data being missing have been observed (they are missing
completely at random meaning the missingness mechanism is completely stochastic). The
cause for missing data will be carefully discussed with the investigator team to determine
whether the missing at random assumption holds. The prediction of missing values will be
performed after specifying a model from the variables distribution (e.g. linear regression
model for normal distribution, logistic model for binary regression…). MICE will be
performed using the R package mice with at least five imputed datasets16.
Analysis of the primary endpoint
1.13 Derivation
Measurement. Under the protocol, participants are scheduled to have an OGTT at week 6
(visit 6, D42+/- 3 days). Plasma glucose at 0 minute will be measured after 12 hours fasting
and before the 75g glucose load via an OGTT.
ASPIRE – SAP – version 3 of October 19th, 2018 18
Data transformation. Available data from Arkadianos et al. suggests that PFG may not be
normally distributed (see Annex)17. In the study by Arkadianos, post intervention PFG
showed significant departure from normality with significant skewness (0.8, Agostinos test; p
= 0.04). By contrast, no significant departure from normality was observed for PFG changes
from baseline (Annex). Although the log transformation has been suggested in the literature,
this transformation applied to Arkadianos data did not improve the normality of PFG
distribution. Using the box-cox method, the best transformation was achieved with the
boundary value of -2.
For this analysis, the preferred approach to handle departures from normality and outliers will
be to use of robust statistical methods18.
1.14 Planned analysis
The primary analysis will be conducted under the ITT principle using modified ANCOVA as
described above. Raw and adjusted mean differences in PFG at week 6 between the
intervention and control arm will be presented with 95% confidence interval (95% CI). The
adjusted mean difference will be estimated using a linear regression model of week 6 FPG to
baseline PFG and treatment arm adjusted for age and sex.
The primary null hypothesis is of no difference in week 6 FPG adjusted means between the
intervention and the control arms. It will be tested against the alternative hypothesis of a
difference between the two groups.
H0: µ0 = µ1
HA: µ0 ≠ µ1
Where µ0 and µ1 are the week adjusted trimmed mean FPG in respectively the control and
intervention group. The null hypothesis will be rejected if the robust adjusted 95% CI of the
mean difference does not include 0.
- Power of the analysis
With 60 participants in each group and assuming that the average PFG changes from baseline
to the week 6 will be -0.6 mmol/l in the intervention arm and -0.1 mmol/l in the control arm,
that the baseline standard deviation (sd) of PFG is 0.7, the post-intervention sd is 0.6, and that
the correlation between baseline and post-intervention measurement range between 0.6 and
0.4, the analysis will have a power greater than 0.9 to detect a difference between both arms.
ASPIRE – SAP – version 3 of October 19th, 2018 19
- Sub-group and additional analysis of the primary outcome
An interaction between baseline FPG and treatment effect will be tested.
No sub-group analysis is planned for the primary endpoint.
- Sensitivity analysis
The robustness of the results will be tested using a non-parametric model as described above.
Both results will be presented and discrepancies will be discussed.
Secondary endpoints
The same analyses (described in the method section and detailed for week 6 FPG above) will
be computed for all of the following continuous variables. Therefore, only additional specific
analyses are described in the section below.
1.15 Variables related to glucose regulation:
Participants are scheduled to have an OGTT at baseline (visit 3), weeks 6 (visit 6), week 12
(visit 8) and week 26 (visit 10).
• Fasting Plasma glucose
- Derivation: plasma glucose after a 12 hours fast and before the 75g glucose load
will be measured via an OGTT.
- Analysis: modified ANCOVA and LMM; no additional analysis.
• Plasma glucose at 120 minutes following 75g OGTT (G120)
- Derivation: Plasma G120 minutes after 75g glucose load will be measured via an
OGTT.
- Analysis: modified ANCOVA and LMM; no additional analysis.
• HbA1c will be measured by the Pathology Department in HammersmithHospital
(Imperial College Healthcare NHS Trust).
- Analysis: modified ANCOVA and LMM; no additional analysis.
• Insulin resistance index computed using the Homeostasis model assessment (HOMA)
of beta cell function
ASPIRE – SAP – version 3 of October 19th, 2018 20
- Derivation: the insulin resistance index (IRI) will be derived using the HOMA
mathematical model19. It requires the FPG and fasting insulin concentration and is
computed as follows:
where glucose stand for FPG in mmol/l and insulin for fasting insulin in mU/l.
- Analysis: modified ANCOVA and LMM; no additional analysis.
• C-peptide at 120 minutes following 75g OGTT will be measured by the Pathology
Department in Hammersmith Hospital (Imperial College Healthcare NHS Trust).
- Analysis: modified ANCOVA and LMM; no additional analysis.
1.16 Variables related to anthropometric measurements
Patients are scheduled to have anthropometric measurements at baseline and visits 4, 6, 8 and
10 (weeks 6, 12 and 26).
• Body weight
- Derivation: weight will be recorded in light clothing without shoes to the nearest
kilogram using the same scale during the trial
- Analysis:
• Main analysis: modified ANCOVA and LMM.
• In addition, changes in weight over time will be analysed using a LMM
model first with measurements taken at weeks 4, 6, 12 and 26.
• Body Mass Index (BMI)
- Derivation: height will be measured bare foot to the nearest 1 cm using a
stadiometer. BMI will be computed as
- Analysis:
• Main analysis: modified ANCOVA and LMM.
• Changes in BMI over time will be analysed using a LMM model first with
only weight measurements taken at weeks 6, 12 and 26 then with all data
including weight reported by phones
• Lean mass
ASPIRE – SAP – version 3 of October 19th, 2018 21
- Derivation: lean mass will be measured using Bioelectrical Impedance Analysis
with the Body Composition Analyser Tanita BC-418
- Analysis: modified ANCOVA and LMM; no additional analysis.
• Fat mass
- Derivation: fat mass will be measured using Bioelectrical Impedance Analysis
with the Body Composition Analyser Tanita BC-418.
- Analysis: modified ANCOVA and LMM; no additional analysis.
• Waist circumference
- Derivation: waist circumference will be measured with to the nearest 0.5 cm at the
midpoint of lower border of the costal margin and uppermost border of iliac crest
using a non-stretchable measuring tape. Three consecutive measurements will be
recorded.
- Analysis: modified ANCOVA and LMM; no additional analysis.
1.17 Variables related to lipid profile
- Derivation: Total and HDL cholesterol and triglycerides will be measured by the
Pathology Department in Hammersmith Hospital (Imperial College Healthcare
NHS Trust). LDL- cholesterol will be computed using the Friedwald relation20:
- Analysis: modified ANCOVA and LMM; no additional analysis.
1.18 Variables related to clinical markers
• Systolic and diastolic blood pressure (BP)
- Derivation: the mean of two measurements of the blood pressure with a digital
sphygmomanometer. Blood pressure will be measured after at least 5’ rest.
- Analysis: the analysis will be conducted separately for systolic and diastolic BP
using modified ANCOVA and LMM as described above; no additional analysis.
1.19 Variables related to nutrition
• Dietary intakes
ASPIRE – SAP – version 3 of October 19th, 2018 22
- Derivation: dietary intakes (Total energy, carbohydrate, fat, saturated fat, salt and
vitamin B6, B12, D) will be assessed using 24-hours recall questionnaire (food
frequency questionnaire [FFQ]) at visits 4, 5, 7, 9 and 11. The FETA software will
be used to derive average daily nutrition contents from the FFQ data. In addition,
25-hydroxyvitamin D and vitamin B6 will be measured directly by the Pathology
Department in Hammersmith Hospital (Imperial College Healthcare NHS Trust).
- Analysis: the analysis will be conducted separately for each nutrient using
modified ANCOVA and LMM as described above;
- Additional analysis: additional dietary intakes may be examined.
Exploratory analyses
The first pre-defined exploratory analysis will compare the exploratory arm with the two
other arms. More specifically it will test the two following hypotheses: The exploratory arm
confers similar or greater benefit compared to the interventional arm and greater benefit
compared to the control arm.
The second hypothesis generating exploratory analysis will examine how the changes in
leptin measurements correlate with changes in weight overall and by intervention arm. Leptin
will be measured via saliva samples at weeks 0, 6, 12 and 26.
The last exploratory analysis will examine how outcomes in the exploratory arm vary across
adherence levels. Adherence levels will be defined from the recorded activity of the
application.
These analyses are exploratory in nature hence results will be interpreted cautiously.
Additional information
1.20 Significance level
All analyses will be conducted using a 5% significance level. No adjustment for multiplicity
will be used.
1.21 Statistical software
Data management will be performed using the statistical software Stata version 12.1 (LP
StataCorp). Statistical analyses will be performed with the statistical software R (R
ASPIRE – SAP – version 3 of October 19th, 2018 23
Foundation for Statistical Computing). The R packages WRS2, robustbase, lme4 and
robustlmm will be used.
ASPIRE – SAP – version 3 of October 19th, 2018 24
References
1. Mathers CD, Loncar D. Projections of global mortality and burden of disease from
2002 to 2030. PLoS medicine 2006; 3(11): e442.
2. WHO. Global report on diabetes. 2016.
http://apps.who.int/iris/bitstream/handle/10665/204871/9789241565257_eng.pdf;sequence=1.
3. Gatineau M, Hancock C, Holman N, et al. Adult obesity and type 2 diabetes. 2014.
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/
file/338934/Adult_obesity_and_type_2_diabetes_.pdf (accessed June 14th, 2018.
4. Diabetes UK. Diabetes: facts and stats. 2014. https://www.diabetes.org.uk/resources-
s3/2017-11/diabetes-key-stats-guidelines-april2014.pdf (accessed June 14th, 2018.
5. Hex N, Bartlett C, Wright D, Taylor M, Varley D. Estimating the current and future
costs of Type 1 and Type 2 diabetes in the UK, including direct health costs and indirect
societal and productivity costs. Diabetic medicine : a journal of the British Diabetic
Association 2012; 29(7): 855-62.
6. Dunkley AJ, Bodicoat DH, Greaves CJ, et al. Diabetes prevention in the real world:
effectiveness of pragmatic lifestyle interventions for the prevention of type 2 diabetes and of
the impact of adherence to guideline recommendations: a systematic review and meta-
analysis. Diabetes care 2014; 37(4): 922-33.
7. Knowler WC, Barrett-Connor E, Fowler SE, et al. Reduction in the incidence of type 2
diabetes with lifestyle intervention or metformin. The New England journal of medicine 2002;
346(6): 393-403.
8. Franks PW, Poveda A. Lifestyle and precision diabetes medicine: will genomics help
optimise the prediction, prevention and treatment of type 2 diabetes through lifestyle therapy?
Diabetologia 2017; 60(5): 784-92.
9. Senn S. Testing for baseline balance in clinical trials. Statistics in Medicine 1994;
13(17): 1715-26.
10. Yang L, Tsiatis A. Efficiency study of estimators for a treatment effect in a pretest –
posttest trial. The American Statistician 2001; 55(5): 314-21.
11. O'Connell N, Dai L, Jiang Y, et al. Methods for analysis of pre-post data in clinical
research: a comparison of five common methods. . J Biol Biostat 2017; 8: 334.
12. Koller. Simulations for Sharpening Wald-type Inference in Robust Regression for
Small Samples. 2018. https://cran.r-
project.org/web/packages/robustbase/vignettes/lmrob_simulation.pdf.
ASPIRE – SAP – version 3 of October 19th, 2018 25
13. Laird N, Ware J. Random-Effects Models for Longitudinal Data Biometrics
1982;38:963-974. Biometrics 1982; 38(4): 963-74.
14. Bates JCPDM, Pinheiro JC, Pinheiro J, Bates D. Mixed-Effects Models in S and S-
PLUS: Springer New York; 2000.
15. Rubin DB. Multiple Imputation for Nonresponse in Surveys: Wiley; 2009.
16. van Buuren S, Groothuis-Oudshoorn K. mice: Multivariate Imputation by Chained
Equations in R. Journal of Statistical Software 2011; 45(3): 1-67.
17. Arkadianos I, Valdes AM, Marinos E, Florou A, Gill RD, Grimaldi KA. Improved
weight management using genetic information to personalize a calorie controlled diet.
Nutrition journal 2007; 6: 29.
18. Wilcox RR. Introduction to Robust Estimation and Hypothesis Testing: Elsevier
Science; 2016.
19. Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF, Turner RC.
Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma
glucose and insulin concentrations in man. Diabetologia 1985; 28(7): 412-9.
20. Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concentration of low-
density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge.
Clinical chemistry 1972; 18(6): 499-502.